This is, in part, to better support shutdown_scripts as part of #4677 and #6139. But improved user experience is also a big factor.

Problem: Even though the coder agent initiates a graceful shutdown, the SSH connections are abruptly closed leaving SSH clients hanging. Similarly any process started inside an SSH session will be left running (e.g. sleep 1000 &) after the SSH server is closed.

To improve upon this, we need to:

• For all active sessions with TTYs, print a message that the workspace is closing
  • Let's defer this, see Refactor coder agent SSH server for better process and connection control #6177 (comment)
• feat(agentssh): Gracefully close SSH sessions on Close #7027
• Send an interrupt (e.g. SIGHUP) to the commands/shells started via SSH
  • Wait for processes to exit (for some amount of time)
  • Sens SIGKILL to the remaining processes
• Stop accepting new SSH connections (potentially allow --debug or similar flag overrides)
• Do not run shutdown_script until the above is complete
• refactor(agent): Move SSH server into agentssh package #7004

I think it would make sense to refactor SSH functionality for the agent as a separate package, e.g. agentssh or agent/sshd. See list above.

Future work/ideas:

• Consider forking/forkexec coder agent to manage new SSH connections
  • This could allow the coder agent to be updated without terminating active connections